Sentiment Analysis API

Table of contents

• Introduction
• Technologies
• Project Folder Structure
• Installation
• Output
• Evaluation

Introduction

Building a backend service that exposes a RESTful API endpoint for sentiment analysis. The API will accept text input and return the sentiment analysis result using a pre-trained model Implemented a web server using Python web framework Flask. Created a single endpoint /analyze that accepts POST requests. Text of user input “Text to be analyzed” for sentiment analysis performed by using a pre-trained machine learning model form Hugging Face Transformers Library. Then the sentiment analysis returns a JSON response a results with structure

{ ‘sentiment’: “Positive/Negative/Neutral” }

Technologies

Project is created with:

• Python version: 3.10.6
• Flask Framework version: 2.2.2
• HuggingFace Sentiment analysis model (pre-trained): https://huggingface.co/StatsGary/setfit-ft-sentinent-eval
• HTML5
• CSS3

Project Folder Structure

Folder name: sentiment_analysis_api Subfolder and files:

-static - css | styles.css

-templates | home.html

app.py

Installation

Instruction to run the project locally (with steps):

---

Open a command prompt and change your directory to the project directory.

step 1:

Set up an virutal environment (venv)

python -m venv venv

You can change the name of your venv as you like.

step 2:

Activate the created virtual environment.

venv/Scripts/activate.bat

This is for windows sytem. For linux/mac

source venv/bin/activate

step 3:

After installing flask. Install transformers.

pip install transformers

step 4:

Install torch in project directory.

pip install flask torch

step 5:

Now, after the convenient codes are written in app.py file set the flask app by writing following command.

set FLASK_APP=app.py

If the file name changes, modify the 'app.py' name with changed name of file.

step 6:

Now, run the flask app by following command.

flask run

Now open up the browser and write http://127.0.0.1:5000 you can modify your port number. Open your app.py file go to app.run() section. then set up like

• app.run(debug=True, port=8080) Then you will need to write in browser: http://127.0.0.1:8080

After all that. If you want to operate that program again using your comand prompt follow (step 2), and (step 6).

Output

Sample Text Input	Sample Sentiment Output
Thank you for the opportunity.	Positive
I disliked the theme.	Positive
I liked the movie but the seats were uncomfortable.	Negative
I like playing football and cricket.	Positive
There is a road by our house.	Positive
I love drinking cold water in the hot summer.	Positive
I enjoyed the food but rude behavior of the staff and disgusting surroundings I did not like.	Negative
I have a good laptop.	Positive
The sky is blue today.	Positive

Evaluation

To evaluate a program it is important to consider both its accuracy and limitations. Sentiment analysis program using Huggin Face Transformers library is able to correctly identify positive and negative sentiment sentiment; this struggle with neutral statements may indicate that it is not effective at distinguishing between ambiguous or mixed emotion.

To Improve the performance of sentiment analysis program:

1. More sophisticated ML algorithms can be used. While the Hugging Face Transformers library is a powerful resource for NLP, it may be good to explore other ML techniques to improve accuracy like CNNs or LSTM models.

2. Fine tuning sentiment analysis program model by modifying hyperparameters or tweaking its architecture. Adjusting number of layers, changing activation function or different learning rates to improve its ability to identify neutral sentiment.

Thank you. :)